An optimized random under sampling boosting model for intrusion detection in wireless sensor network

The rapid growth of Information Technology and the COVID-19 pandemic have significantly increased internet usage worldwide, leading to a surge in cyber threats. Network Intrusion Detection Systems (NIDS) are essential for monitoring and mitigating unauthorized intrusions, but class imbalance among attack types often hampers performance, particularly for minority samples. This research proposes an Optimized Random Under sampling Boosting Model using Particle Swarm Optimization algorithm to enhance the detection and classification of five intrusion types: DOS, Probe, R2L, U2R, and Normal. The model outperformed state-of-the-art models such as JN8, KNN, and GA–LSTM–RNA, achieving near-perfect performance with approximately 100% accuracy for all attack types. In terms of precision, the model achieved 99.990% for DOS, 99.886% for Probe, 97.689% for R2L, 73.684% for U2R, and 99.990% for Normal. The recall rates were similarly impressive, with 99.993% for DOS, 99.943% for Probe, 99.329% for R2L, 93.333% for U2R, and 99.886% for Normal. The F1-scores were 99.956% for DOS, 99.914% for Probe, 98.502% for R2L, 82.353% for U2R, and 99.938% for Normal. In general, the proposed ORUSBEM model was able to effectively detect each type of attack, including the normal type. The results revealed the effectiveness of the developed technique in enhancing the accuracy and efficiency of the intrusion detection system. The result is beneficiary in improving the security of various applications by detecting and preventing malicious attacks. The model executed 100 iterations in 185.1 seconds, indicating significant optimization improvements. While BPSO corroborate to be effective in selecting the optimal feature subset, other optimization techniques such as Genetic Algorithms (GA), Ant Colony Optimization (ACO), or Differential Evolution (DE) can be used to compare and optimize the results in the future. However, focusing on enhancing U2R detection and reducing execution time is also an area that needs serious attention. Additionally, testing the model in real-life applications is crucial for strengthening cyber security in industries and organizations.